LongCat 2.0 Pricing Explained: API Cost, Cache Hits & Is It Worth It?
Understand LongCat 2.0 pricing, uncached and cached input, output tokens, LongCat 2.0 Online credits, plans, and cost-per-task decisions.
LongCat 2.0 Online charges for Playground and API use through credits. The total depends on prompt size, output length, cache behavior, and how often an agent retries a task. A reasonable credit rate does not prevent waste when an application repeatedly sends a large repository or lets a coding loop run without limits.
Direct answer: LongCat 2.0 Online currently offers monthly plans from $19 and one-time credit packs from $40. New accounts receive 2,000 evaluation credits. The live pricing page is the source of truth for customer charges and was last reviewed for this article on July 10, 2026.
LongCat 2.0 Online prices at a glance
| Option | Price | Included credits |
|---|---|---|
| Starter monthly | $19/month | 22,000 |
| Plus monthly | $29/month | 40,000 |
| Pro monthly | $59/month | 95,000 |
| Top-up pack | $40 once | 35,000 |
| Growth pack | $80 once | 80,000 |
| Scale pack | $150 once | 160,000 |
These are LongCat 2.0 Online retail prices. The service includes the hosted Playground, account credits, API-key management, usage metering, billing integration, and service operation. Confirm the current amount and validity period on the pricing page before purchase.
What uncached input means
Uncached input is prompt content the provider processes without a reusable cache hit. It can include system instructions, conversation history, source code, retrieved documents, tool results, and the latest user message.
A coding agent often resends stable instructions and repository context across turns. If every request is uncached, that repeated material is billed at the full input rate. LongCat 2.0's one-million-token context makes large inputs possible, but possible is not the same as economical.
Keep prompts focused. Send a repository map and relevant files before sending the entire tree. Remove duplicate logs, generated files, vendored code, and stale documentation.
What a cache hit changes
Cached input uses fewer site credits when the provider recognizes reusable prompt content.
That discount can matter for:
- A stable system prompt used across many requests.
- Shared repository context during an agent session.
- Long reference documents reused for several questions.
- Repeated coding instructions and tool definitions.
Do not assume a cache hit merely because two prompts look similar. Cache rules can depend on exact prefixes, provider settings, cache lifetime, and request format. Check the usage object or billing dashboard for the provider's confirmed cached-token count.
Why output often controls the bill
Output tokens cost more than uncached input tokens. Coding agents can generate long explanations, patches, tool calls, and repeated corrections. A request with a short prompt can still become expensive if the model produces thousands of tokens over several turns.
Set a reasonable max_tokens value, ask for concise plans, and stop an agent that repeats the same failed action. Measure cost per completed task rather than cost per individual request.
How LongCat 2.0 Online credits work
LongCat 2.0 Online provides a browser Playground and an OpenAI-compatible API. Both use account credits rather than charging a card for each raw upstream request.
Current metering rules are:
| Usage | LongCat 2.0 Online credits per 1M tokens |
|---|---|
| Uncached input | 900 credits |
| Cached input | 20 credits |
| Output | 3,600 credits |
The system calculates credits from actual usage returned after a successful generation and rounds a successful request up to at least one credit. Before generation, it estimates whether the account has enough balance for the allowed output. A successful response creates a usage transaction. Invalid authentication, rejected requests, and upstream failures are not successful generations.
Free evaluation credits
Registration grants 2,000 credits. A daily login grants 100 credits, capped at 1,000 login credits per month. Gift credits expire after 30 days.
This makes the LongCat 2.0 Online Playground suitable for evaluation without offering anonymous, unlimited access. Login ties rate limits and usage to an account and reduces abuse that would otherwise create an uncontrolled API bill.
Monthly plans and credit packs
The site's pricing page is the authoritative source for current retail plans. At publication time, the product configuration includes monthly Starter, Plus, and Pro plans, plus Top-up, Growth, and Scale credit packs.
Monthly plans are designed for recurring Playground and API use. Credit packs are useful when usage is irregular or when a subscriber needs additional capacity. Check the displayed validity period and current Stripe checkout price before purchasing.
Example cost estimates in credits
Short coding question
Assume 1,000 uncached input tokens and 800 output tokens:
input: 1,000 / 1,000,000 × 900 = 0.9 credits
output: 800 / 1,000,000 × 3,600 = 2.88 credits
rounded total: 4 creditsMedium repository review
Assume 50,000 uncached input tokens and 4,000 output tokens:
input: 50,000 / 1,000,000 × 900 = 45 credits
output: 4,000 / 1,000,000 × 3,600 = 14.4 credits
rounded total: 60 creditsRepeated review with cached context
Assume 45,000 cached input tokens, 5,000 new input tokens, and 2,000 output tokens:
cached input: 45,000 / 1,000,000 × 20 = 0.9 credits
new input: 5,000 / 1,000,000 × 900 = 4.5 credits
output: 2,000 / 1,000,000 × 3,600 = 7.2 credits
rounded total: 13 creditsThese examples explain the formula. Actual usage depends on tokenization and the provider's usage response.
Is LongCat 2.0 worth the price?
LongCat 2.0 is worth testing when you need agentic coding, long context, or lower API rates. It is especially attractive when a repeated context receives real cache hits. It may not be the best value if a more expensive model completes difficult tasks with far fewer corrections.
Use a small evaluation set and measure:
- Tasks completed without manual repair.
- Input and output tokens per completed task.
- Latency and retry count.
- Quality of tests and code review.
- Total credits or provider cost.
Start with the online Playground. When the output meets your requirements, create a key using the LongCat 2.0 API guide and automate the same prompt. The pricing page should be checked immediately before purchase because plans and provider costs can change.